COMPARISON OF FREQUENCY-BAND SELECTION STRATEGIES FOR 2D TIME-DOMAIN ACOUSTIC WAVEFORM INVERSION

Full waveform inversion (FWI) is a promising model-building technology to recover subsurface information. However, it is easy to fall into local minima when applying this method because of the lack of a low-frequency component in seismic data. To mitigate this problem, multi-scale method for the time domain has been proposed. With this method, we perform the inversion sequentially from low- to high-frequency groups, and we set the velocity model inverted at the previous scale as an initial velocity model for the next higher frequency group. In this study, we mainly compare several frequency-band selection strategies for FWI in time domain, including individual-grouping methods 1 and 2, along with Bunks' method. To verify and compare the efficiency of the above three methods, we introduce the partial-overlap and arbitrary-two grouping methods. Numerical examples for synthetic data of the Marmousi velocity model, as well as noisy data, demonstrate that multiscale inversion can attain encouraging resolution. Compared to solutions from other methods, we highlight the individual-grouping method 2 which can yield a more satisfactory velocity model. Also, numerical results imply that low frequencies are necessary in full waveform inversion.